Received July 1, 2019, accepted July 17, 2019, date of publication July 22, 2019, date of current version

August 7, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2930237

Robust Adaptive Neural-Network Backstepping

Control Design for High-Speed Permanent-
Magnet Synchronous Motor Drives:
Theory and Experiments
FAYEZ F. M. EL-SOUSY 1,4 , (Member, IEEE), MOHAMED F. EL-NAGGAR1,2 ,
MAHMOUD AMIN 3,4 , (Senior Member, IEEE),
AHMED ABU-SIADA 5 , (Senior Member, IEEE), AND KHALED A. ABUHASEL 6
1 Electrical Engineering Department, College of Engineering, Prince Sattam Bin Abdulaziz University, Al-Kharj 11942, Saudi Arabia
2 Faculty of Engineering, Electrical Engineering Department, Helwan University, Helwan 11795, Egypt
3 Electrical and Computer Engineering Department, Manhattan College, New York, NY 10471, USA
4 Department of Power Electronics and Energy Conversion, Electronics Research Institute, Giza 12622, Egypt
5 Electrical and Computer Engineering Department, Curtin University, Perth, WA 6845, Australia
6 Mechanical Engineering Department, College of Engineering, University of Bisha, Aseer 61421, Saudi Arabia

Corresponding author: Mahmoud Amin (mahmoud.amin@manhattan.edu)

This work has been supported by Deanship of Scientific Research, Prince Sattam bin Abdulaziz University, Saudi Arabia, Under Grant
Number (2017/ 01/ 7126).

ABSTRACT This paper presents a robust adaptive backstepping control (RABC) for high-speed
permanent-magnet synchronous motor (HSPMSM) drive system. The proposed RABC achieves high
performance operation by incorporating an ideal backstepping controller (IBC), a recurrent radial basis
function neural network (RRBFNN) uncertainty observer, and a robust controller. The Lyapunov stability
theorem is utilized to design the IBC as a position controller of the HSPMSM servo drive system. To enhance
the disturbance rejection capability during parameter changes, certain information is needed within the
backstepping control law so that the system performance would not sorely be affected. To mitigate the need
for the lumped parameter uncertainties within the backstepping controller, an online adaptive observer based
on RRBFNN is designed to estimate the nonlinear parameter uncertainties. Moreover, the robust controller
is intended to retrieve the remaining of the RRBFNN approximation errors. To assure the stability of the
proposed RABC, the Lyapunov stability analysis is used to derive the online adaptive control laws. The
performance of the proposed RABC is verified by simulation and experimental analysis under different
operating conditions and parameter uncertainties. The test results validate the effectiveness of the proposed
RABC scheme to achieve preferable tracking performance regardless of external disturbances and parameter
uncertainties.

INDEX TERMS Adaptive control, backstepping technique, Lyapunov stability theorem, high-speed
permanent-magnet synchronous motor, radial basis function neural network (RBFNN), uncertainty observer.

I. INTRODUCTION motors (micro PMSMs) provide high efficiency, robustness,

In recent years, several processing techniques of micro-
electromechanical systems (MEMS) have been developed
to reduce power dissipation, size, and weight of the micro-
motors. For special industrial applications, micromotors
are considered good candidates to achieve high perfor-
mance operation. The micro permanent-magnet synchronous
The associate editor coordinating the review of this manuscript and
approving it for publication was Gaolin Wang.
high power density, better reliability, and high speed oper-
ation compared to other micromotors [2], [3]. Furthermore,
micro PMSMs are good nominees for several industrial appli-
cations such as medical diagnostic, surgical devices, security
equipment, power driving devices in MEMS, and micro auto-
nomic robots [4].
In the literature, several topologies and control techniques
have been developed for micromotors [5]–[17]. In [5], a com-
parative analysis of inverter topologies for micromotors has

VOLUME 7, 2019 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ 99327
 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design
been performed. A six-phase five-level inverter prototype was
designed and tested at high frequency of 2 MHz. Various
speed sensorless control techniques have been introduced
for the micromotor drives [6]–[12]. For sensorless control of
micro PMSMs, the rotor position measurement is estimated in
the control schemes [6], [7]. A micro PMSM control system
based on rotor position estimator and robust H∞ controller
was introduced in [8]–[10]. Several advanced control strate-
gies for micromotors have been studied in [11] and [12].
For micro PMSM system in [13], a robust identifier and
intelligent Petri-fuzzy neural network (PFNN) controller was
proposed. An adaptive inverse control scheme incorporates
an adaptive model for a micro PMSM drive system was
developed [14]. In [15], an optimal design of sensorless-
based speed controller of micro PMSM drives was proposed.
Moreover, a motion controller involves a tuning parame-
ter feed-forward function along with an optimal position
controller was implemented for micro PMSM system [16].
In [17], a sliding-mode observer has been incorporated in
a robust sensorless control method for high-speed micro
PMSM.
Since micro PMSM drive system is extremely nonlinear,
uncertain and has a wide range of operating conditions,
the linearization around one operating point cannot be used
to design the controller. To resolve this issue, nonlinear
control techniques can be effectively utilized [18]–[21]. The
recent development of backstepping control technique is a
robust and consistent design approach for nonlinear feed-
back control schemes, which provides an option to conform
the unmodeled and nonlinear impacts and parameter uncer-
tainties. Several backstepping control design approaches
have been suggested to handle nonlinear systems and
motor drives [22]–[26]. The adaptive backstepping control
approach is designed based on choosing recursively several
proper functions of state variables as pseudo control inputs
with reduced order subsystems of the whole control scheme.
A new pseudo control structure from preceding stages is
generated for each backstepping stage. The summation of all
Lyapunov functions resulted from each design stage produces
final Lyapunov function which fulfills the main design objec-
tive while terminating the actual control input of the feed-
back design. Therefore, favorable robustness features under
parameter uncertainties are achieved using the backstepping
control approach [18], [23]–[25].
Intelligent computation methodologies are attracting more
attention due to rapid industrial developments. These method-
ologies have been evolved to enhance the drive system
characteristics and to address the uncertainties and nonlinear-
ities [27]–[34]. The architecture of the radial-basis function
neural network (RBFNN) is designed based on the Kohonen
network model. The RBFNN is structured of input, hidden,
and output layers with normalized Gaussian activation func-
tions. Although the simple structure of the RBFNN, it is
one of the most preferable observers for position/speed con-
trollers and nonlinear mapping problems due to its superior
performance [35]–[40]. RBFNN has a quicker convergence
99328
feature than standard multilayer-perception neural network.
RBFNN has a comparable characteristic as the fuzzy-logic
scheme, in which the output result is studied employing
the weighted-sum technique, and the number of nodes in
the hidden layer is similar the fuzzy system structured of
‘‘if-then’’ rules. Furthermore, the RBFNN is represented
by the field function which is the same as that of the
fuzzy-logic system with its premise part constructed of the
membership functions. Thus, the RBFNN can be incor-
porated effectively with nonlinear controllers designed for
dynamic systems with parameter uncertainties since it intro-
duces several features of its self-adaptation characteristics
and numerous facets [40]. In [41], an improved performance
of the shunt active power filter (APF) is accomplished by
developing a radial basis function neural network (RBFNN)
incorporated in an adaptive fuzzy-neural-network schematic.
The proposed RBFNN is used to enhance the dynamic model
of the APF through approximating its nonlinear function.
Furthermore, the Lyapunov stability analysis is utilized to
develop an online adaptive law in order to adjust the weights
of the proposed RBFNN. In [42], a nonsingular terminal slid-
ing mode backstepping (NTSMB) control approach is pro-
posed to design an adaptive fuzzy-neural-network (AFNN)
to mitigate the influence of the APF dynamic model uncer-
tainties and external perturbations. In addition, the NTSMB
robustness is enhanced by relieving the need of the preceding
information of system specifications. In [43], the perfor-
mance of the photovoltaic (PV)-based grid-connected single-
phase inverter is improved using disturbance observer-based
fuzzy sliding mode control (DOBFSMC) scheme. The pro-
posed observer is utilized as an online estimator of system
disturbances while the inverter output voltage is controlled
using a sliding mode controller based on the information gen-
erated by the observer. Meanwhile, the system performance
is enhanced through approximating the error value of the
observer using fuzzy control system. Moreover, a terminal
sliding-mode-based adaptive current controller for an APF
is proposed [44]. To assure stable sliding surface properties
with high accuracy, an adaptive finite-time fractional-order
control scheme is designed. A fuzzy-neural observer is con-
structed to estimate the unknown nonlinearities of the APF
while suppressing the current harmonic distortion. In [45],
a robust adaptive vibration control is desired for systems with
flexible risers subjected to input nonlinearities and unknown
external disturbances. To eliminate the influence of input non-
linearities and limit the vibrational offset, a robust adaptive
boundary controller is developed. Furthermore, the vibration
control approach along with the adaptive upper-bound law
is used to estimate the unknown disturbance boundary mag-
nitude. In [46], another adaptive neural network based on
backstepping approach is proposed to control a vibrating flex-
ible string system. The effect of system uncertainties, input
asymmetrical dead-region, and output restraint are consid-
ered. The proposed backstepping control strategy is designed
to assure that the output constraints are not overridden. Mean-
while, the neural network is constructed to recover the input
VOLUME 7, 2019
F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design
FIGURE 1. The proposed RABC framework for the HSPMSM drive system.
asymmetrical dead-region effect and maintain the overall
string system stability. In [47], a new adaptive boundary
control system is proposed to suppress the vibration of a belt
system with axial movement and compensate for parametric
disturbances. In addition, a disturbance observe is designed to
mitigate the influence of the unknown boundary uncertain-
ties. In [48], another boundary control scheme is desired to
reduce the vibration of the flexible string system under the
effect of external perturbations and input dead-region. The
proposed scheme employs the backstepping control strategy
to mitigate the vibration of the string system. Subsequently,
the influence of the input dead-region is resolved using a
RBFNN while the external perturbations are handled utilizing
a disturbance observer.
This main contribution of this paper is to design a RABC
scheme for the HSPMSM servo drive using an intelligent
adaptive backstepping control system. The drive system
structure of the designed RABC is demonstrated in Fig. 1.
For industrial applications, the compounded disturbances and
accurate lumped parameter uncertainties are harsh to be iden-
tified beforehand of the HSPMSM control operation. Thus,
a novel scheme of nonlinear controller is considered here
based on the adaptive backstepping control with RRBFNN to
achieve the desired performance. Accordingly, the proposed
control scheme incorporates three parts: an IBC, a RRBFNN-
based uncertainty observer and a robust controller. Utilizing
Lyapunov stability theorem, an effective design of the pro-
posed RABC scheme is established to accurately control the
rotor position of the HSPMSM drive system. Meanwhile,
the uncertainty term existed in the backstepping control law
is needed to be recognized as a mean to reduce the severe
impact of the parameter changes on the system performance.
The IBC can effectively control, track and regulate the rotor
position of the drive system. Though, the drive system is still
significantly affected by virtue of the existing uncertainties
VOLUME 7, 2019
involving the unexpected disturbances, parameter variations,
and inevitable approximation errors. This problem can be
solved by using a RRBFNN uncertainty online observer to
adaptively estimate the nonlinear parameter uncertainties.
Furthermore, a robust controller is designed to retrieve the
remaining of the relative errors of the RRBFNN. The Lya-
punov stability analysis is used to assure the closed-loop
system stability theory. The validity of the proposed RABC
design is confirmed by test results (simulation and exper-
imentation) subject tooled variations and parameter uncer-
tainties. The test outcomes assure the effectiveness of the
proposed RABC design through eliminating the external
load perturbations in addition to compensating the parameter
uncertainties.
Finally, the contributions of the proposed RABC scheme
compared to other schemes are concluded as:
• The RRBFNN model has a new structure with the
advantage of recurrent property which handles interim
issues.
• The RRBFNN model has a quicker convergence feature
than standard multilayer-perception NN.
• The RRBFNN model has a comparable characteristic
as the fuzzy-logic scheme, in which the output result is
studied employing the weighted-sum technique,
• The RRBFNN model has a hidden layer with num-
ber of nodes similar to the fuzzy system structured of
‘‘IF-THEN’’ rules.
• The RRBFNN model is represented by the field function
which is the same as that of the fuzzy-logic system
with its premise part constructed of the membership
functions.
• The feedback of the output layer is added to accomplish
faster convergence time.
• The RRBFNN neurons are susceptible to past data due
to the self-connections of the hidden.
99329
 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

• Compared with RBFN, faster convergence and higher The HSPMSM motion dynamic equation can be described as:
precision are achieved.
d
• In Section IV, it was validated that the proposed Te − TL = Jm (2/Pn ) ωr + βm (2/Pn ) ωr (4)
RABC with the RRBFNN uncertainty observer main- dt
tains robust control features and effectively controls where Vqs , Vds , iqs and ids are the d − s tator voltages and cur-
the HSPMSM system under multiple perturbations and rents. Rs and Lss are the stator resistance and self-inductance.
parameter uncertainties. θr , ωr , Jm , βm and P are the rotor position, electrical rotor
This paper is structured as: Section II introduces the speed, effective inertia, friction coefficient and the number
HSPMSM dynamic modeling with parameter uncertainty and of poles of the HSPMSM, respectively. TL and Te are the
the problem description. In section III, the design proce- load and electromagnetic torques, respectively. The torque
dure of the RABC scheme is presented. First, the structure constant is expressed as Kt = (3/2)(Pn /2) · λ0m .
of the RRBFNN uncertainty observer is provided. Then, It is common knowledge that the FOC of the HSPMSM
the detailed design steps for the proposed IBC are also pre- enables an independent control of two input state variables,
sented in this section. Moreover, the adaptive training meth- stator d − q-axis currents irds and irqs . The dynamic model of
ods and the stability study of the designed RABC scheme the HSPMSM (1)-(4) in reliance on the field-oriented control
are illustrated in Section III. A development control board in the synchronous reference frame [18] can be illustrated in
(dSPACE DS1102 DSP) is utilized to implement the proposed state form as:
control algorithms. The HSPMSM servo system has been
θ̇r = ωr

studied to examine the dynamic performance under two dif- 
βm

 Kt r 1
ω̇r = Jm iqs − Jm TL − Jm ωr

ferent conditions (extrinsic load perturbations and parameter 


uncertainties). Section IV provides the test results to confirm
Rs r 1 1 r (5)
i̇qs = − L iqs − ωr ids − L ωr λm + L Vqs
the validity of the proposed RABC design for the HSPMSM  r r 0

servo system. Finally, Section V introduces conclusions and


 ss ss ss
i̇ = − Rs ir + ωr irqs + 1 V r

r

summarizes the main contributions. ds
Lss ds Lss ds
TABLE 1. The three-phase HSPMSM model parameters.
Due to temperature change, load disturbance, and sat-
uration, the motor parameters are changing during motor
operation. Thus, all these possible uncertainty factors should
be considered during the design phase of the drive system
controller. Accordingly, the perturbed dynamic model of the
previous motor equations presented in (5) can be derived and
expressed by (6)-(15) as follows:

θ̇r = ωr



(Jm + 1Jm )ω̇r = (Kt + 1Kt )irqs − (TL + 1TL )





−(βm + 1βm )ωr


(6)
ss + 1Lss )i̇qs = −(Rs + 1Rs )iqs − ωr ids
 (L r r r


II. PROBLEM FORMULATION AND −ωr λ0m + Vqs r




MATHEMATICAL PRELIMINARIES (Lss + 1Lss )i̇rds = −(Rs + 1Rs )irds + ωr irqs + Vds

 r
A. THE HIGH-SPEED PMSM DYNAMIC 
MODEL WITH UNCERTAINTY


 θ̇r = ωr
ω̇ = Kt irqs −(TL + 1TL )−βm ωr

The field-oriented control (FOC) approach is applied with
 J
 m r

 
+ fr − 1Jm ω̇r − 1βm ωr + 1Kt iqs −1TL −TL
the aim of achieving high torque capability of the HSPMSM


 r
system by virtue of decoupling the d − q axes stator currents

Lssi̇rqs = −Rs irqs −ωr irds − ωr λ0m + Vqsr (7)
in the rotor reference frame [18]. The motor parameters are  
+ fqs − 1Lss i̇qs − 1Rs iqs + 1Lss ωr ids
r r r

denoted in Table 1. The analytical modeling of the HSPMSM




in the rotating reference frame can be represented as: Lss i̇rds = −Rs irds + ωr irqs + Vds

 r



d r + f − 1L i̇r − 1R ir + 1L ω ir


r
Vqs = Rs irqs + Lss i + ωr Lss irds + ωr λ0m (1) ds ss ds s ds ss r ds
dt qs
d The state vector and its derivative are given by:
r
Vds = Rs irds + Lss irds − ωr Lss irqs (2)
dt θ̇r
θr
       
x1 ẋ1
The electromagnetic torque can be represented by:  x2   ωr   ẋ2   ω̇r 
x=  x3  =  irqs  , ẋ =  ẋ3  =  i̇rqs  (8)
      
3 Pn 0 r
Te = · · λm · iqs = Kt irqs (3)
2 2 x4 irds ẋ4 i̇rds

99330 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

Using (6)-(8), the perturbed dynamical model of the of uncertainty cannot be directly evaluated. Subsequently, the
HSPMSM is expressed in the following form: RRBFNN-based uncertainty observer is proposed to estimate
the nonlinear lumped parameter uncertainty terms for the
ẋ1 = θ̇r = x2 (9)
βm Kt HSPMSM drive system. The control objective is to develop
ẋ2 = ω̇r = − x2 + x3 + r (10) a robust adaptive backstepping control (RABC) system with
Jm Jm RRBFNN uncertainty observer such that the closed-loop sys-
r
Rs λ0 Vqs
tem of (9)-(12) is stable in the existence of parameter uncer-
ẋ3 = i̇rqs = − x3 −x2 x4 − m x2 + + qs (11)
Lss Lss Lss tainties and extrinsic perturbations. Eventually, all errors are
Rs Vr consistently restricted and the tracking error value can be
ẋ4 = i̇rds = − x4 + x2 x3 + ds + ds (12)
Lss Lss arbitrary small as t → ∞ . The configuration of the designed
and the uncertainty terms in (10)-(12) are given by (13)-(15): RABC is shown in Fig. 1.
1 In this paper, a novel RABC scheme is designed to control
r = (fr −1Jm ẋ2 −1βm x2 +1Kt x3 −1TL −TL ) (13) the HSPMSM drive system. The proposed scheme integrates
Jm
IBC, RRBFNN-based uncertainty observer and robust con-
1
qs = fqs −1Lss ẋ3 − 1Rs x3 +1Lss x2 x4


(14) troller. First, the mathematical model of the HSPMSM with
Lss parameter variations and external disturbances is derived.
1
ds = (fds −1Lss ẋ4 − 1Rs x4 +1Lss x2 x4 ) (15) The FOC approach is utilized here to enhance the dynamic
Lss performance of the drive system through the decoupling con-
where r , qs and ds represent the lumped parameter trol property. In accordance with the backstepping process,
uncertainties, wherein 1Rs , 1Lss , 1Kt , 1βm , 1Jm and 1TL an IBC is designed based on Lyapunov stability theorem
denote possible uncertainties in the drive system parameters; to fulfill several goals of a persistent rotor position while
fr , fqs and fds are added to indicate the extrinsic perturbations tracing the desired trajectory. Though, accurate data about
in realistic applications. The lumped parameter limits  ofthe the lumped parameter uncertainties of the drive system are
possible uncertainties are expressed by |r | < δr , qs  <
δqs and |ds | < δds ; δr , δqs and δds are positive constants.
Remark 1: In the dynamic modeling of the motor formu-
lated by (9)-(15) encompasses all uncertainties, the presence
of nonlinearities is explicit due to the d − q stator current,
the compounded rotor speed, and the permanent flux terms.
Furthermore, there occur irregularities as a result of the non-
linear features of the CRPWM inverter. This is a substan-
tial cause of the design difficulty for the robust control of
HSPMSM system. Besides, the parameter variations enlarge
the nonlinearities and reduce the system performance or even
demolish the control stability. To design a superior con-
trol of the motor drive system, particular factors such as
the system nonlinearities, parameter changes and extrinsic
load perturbations should be canceled or restricted within
an attenuation level. Favorably, the whole nonlinear reliance
might be intended into the parameter uncertainties (13)-(15)
and the proposed controller need to be designed robust ade-
quate to with stand un-modeled dynamics as well as these
uncertainties.
Assumption 1: The HSPMSM states, x1 = θr , x2 = ωr ,
x3 = irqs and x4 = irds , are measurable since the backstepping
control scheme requires these feedback signals.
B. PROBLEM DESCRIPTION
The proposed RABC framework for the HSPMSM system
is shown in Fig. 1. The drive system contains a HSPMSM,
a three-phase current regulated pulse width modulation
(CRPWM) inverter and the load. The system uncertainties
exist in the perturbed dynamic model of the HSPMSM drive
system (9)-(12) are presumed to be limited. These uncertain-
ties include extrinsic perturbations as well as unknown mod-
eling inaccuracies. From (13)-(15), it is noticed that the terms
VOLUME 7, 2019
needed within the backstepping control law in such a way
the performance would not sorely affected. To mitigate the
need for the parameter uncertainties within the IBC structure,
an online adaptive observer based on RRBFNN is intended to
evaluate the nonlinear parameter uncertainties. Furthermore,
the robust controller is configured to retrieve the remaining
of the RRBFNN estimate error. To ensure the stability of the
proposed RABC, the Lyapunov stability analysis is employed
to obtain the online adaptive control laws.
III. ROBUST ADAPTIVE BACKSTEPPING CONTROL
VIA RRBFNN UNCERTAINTY OBSERVER
The main idea of the backstepping process is to structure a
new subsystem using preselected state variable that needed
to be stabilized. Therefore, the error functions of the new
state variables are chosen to be attenuated to zero. As a
result, a virtual control law is derived by selecting a proper
Lyapunov candidate function. Eventually, an actual control
law could be concluded, and the proposed system stabil-
ity would be assured. In this section, the structure of the
designed RABC for the HSPMSM is presented. At the nom-
inal parameters, the IBC can fulfill a desirable performance
of the HSPMSM position control. Nevertheless, the control
performance of the HSPMSM is still susceptible to parameter
changes. To resolve this issue with an effective control design
of the rotor position of the HSPMSM, a RABC is developed
here. The structure of the designed RABC system, which
incorporates an IBC, a RRBFNN uncertainty observer and
a robust controller, is shown in Fig. 1.
A. RRBFNN UNCERTAINTY OBSERVER
The HSPMSM drive system involving parameter uncertain-
ties is expressed in terms of the unknown nonlinear parameter
99331

FIGURE 2. Recurrent radial basis function neural network (RRBFNN).
uncertainty function which can be estimated here using the
proposed RRBFNN uncertainty observer. To adaptively esti-
mate this nonlinear dynamic function, (·),
ˆ the backpropaga-
tion algorithm is utilized to train the RRBFNN. To implement
the proposed uncertainty observer, a three-layer RRBFNN
is suggested here to optimize the precision of the function
approximation. The proposed RRBFNN encompasses three-
layers, two inputs (the i layer), hidden (the j layer), and
one output (the o layer), as shown in Fig. 2. In addition,
the Gaussian function is selected as the particular field func-
tion in the hidden layer due to its differential and persistent
characteristics. Furthermore, the RRBFNN output is repeated
to the input of the output layer with a time delay. The basic
function and signal propagation in each layer of the proposed
RRBFNN are expressed as shown below [18]:
Layer 1: Input Layer
Layer 1 involves some nodes to transmit the input signals
to the subsequent layer. In the input layer, every input and
output node i of the RRBFNN can be expressed as:
neti1 = xi1 (N )
y1i (N ) fi1 (neti1 (N )) neti1 (N ), i = 1, · · · , m
= =
F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design
deviation and mean of the jth neuron in the hidden layer
of the ith input of the RRBFNN, n indicates the number
of receptive field units and bases in the hidden layer, X =
[x1 , x2 , . . . , xi ]T ∈ <m×1 is the input vector of the input layer,
8 = [81 , 82 , . . . , 8j , 8n ]T ∈ <n×1 is the output vector of
the hidden layer, 0 ≤ α < 1 is the self-connecting feedback
gain of the hidden layer.
Layer 3: Output Layer P
The single node o in the output layer is represented by ,
which calculates the total output as the addition of all arriving
signals to acquire the final outcomes.
X
neto3 = Wj y2j (N ) + Wo ylo (20)
j
y3o (N ) = fo (neto3 (N ))
3
= neto3 (N )o = 1 (21)
 
8j (N ) = exp −(X−µj ) σj (X−µj )+8j (N −1)αj
T
(22)
ylo (N ) = ylo (N − 1)βj + y3o (N − 1) (23)
y3o = (·)
ˆ (24)
where Wj = [$1 , $2 , . . . , $j , $n ]T ∈ <n×1 is the adaptable
weight vector between the hidden and the output layers, Wo =
[$o ]T ∈ <1×1 is the connective weight of output feedback
neuron to output neuron, ylo is the feedback layer output,
y3o = (·)
ˆ is the output of the RRBFNN, 0 ≤ β < 1 is
the self-connecting feedback gain of the output layer, (·) ˆ
is the function of the nonlinear parameter uncertainty. For
the uncertainty estimation, the vector form of the RRBFNN
output can be expressed as:
y3o (x, σ, µ, α, β, W ) = W T 8(x, σ, µ, α, β) (25)
where W is the weight vector and 8 is the firing strength
vector. Even for time-varying function, it was confirmed
that an RRBFNN exists as shown in (25) such that it can
symmetrically approximate function nonlinearity [40].
Assumption 2: consider the RRBFNN input, X =
[x1 , x2 , . . . , xi ]T , pertains to a compact set κX and the
RRBFNN is utilized to estimate the nonlinear function (X ).
The optimal parameter vector of the proposed RRBFNN, W ∗ ,
(16) is provided by [51]–[54]:
" #
(17)


∗
Wk = arg min sup
k (X ) − 
ˆ k (X Ŵk )
(26)


where xi1 illustrates the ith input to the node of layer 1 and N
indicates the number of iterations.
Layer 2: Hidden Layer
Every node in the hidden layer precedes a susceptible field
function. The Gaussian function is chosen as the receptive
field function in the hidden layer due to its differential and
persistent features. For the jth hidden node:
netj2 (N ) = −(X−µj )T σj (X−µj ) + 8j (N − 1)αj
y2j (N ) = fj2 (netj2 (N )) = exp(netj2 (N )), j = 1, ..., n
where the standard deviation and mean vectors of the Gaus-
sian functions are σj = [1/σ1j2 , 1/σ2j2 , . . . , 1/σij2 ]T and µj =
[µ1j , µ2j , . . . , µij ]T , respectively, µij and σij are the standard
99332
Ŵ ∈κW X ∈κX
It is presumed that the optimal parameter vector, Wk∗ , is
limited to a compact set of the parameter vector, κW .
B. ONLINE LEARNING ALGORITHM
The parameter learning algorithm is used to adjust the param-
eters of the RRBFNN bases, feedback weight and the con-
(18) nection weight (µ, σ , α, βWj and Wo ) optimally with the
(19) same training pattern. The detailed derivation of the learn-
ing methodology is given in the Appendix. In this paper,
j j
the weights, Wj , Wo , αi , and βol as well as the mean µi , and
j
the standard deviation σi , are under training to get the adapted
laws using Lyapunov stability. The weighting vector W which
VOLUME 7, 2019
F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

collects all weights for training, is defined as weight, and estimation errors. The parameter uncertainties in
(13)-(15) can be illustrated as:
W = [Wj , Wo , α, β, µ, σ ]
k (xk Wk∗ ) = ∗k (xk Wk∗ )+εk
 
(w11 , · · · , wnm 1 , . . . , w12 , . . . , wnm 2 , · · · ,
 
 h  i
=ˆ k (xk Ŵk )+ ∗k (xk Wk∗ )− 

 w1nj , · · · , wnm nj ),
 
 ˆ k (xk Ŵk )
 (w1l , · · · , wn 1l , . . . , w2l , . . . , wn 2l , · · · , 
 
 m m  +εk + uRC
k (29)
 w1nl , · · · , wnm nl ),
 
where Ŵk and Wk∗ are the estimated and optimal weight

 (α1j , α2j , . . . , αmj , . . . , αn j ),
 
matrices, respectively. xr = (x1 , x2 , ẋ1 ), xqs = (x1 , x2 , x3 , ẋ2 )

m
=
 
 (β1l , β2l , . . . , βml , . . . , βnm l ), and xds = (x1 , x2 , x3 , ẋ3 ) are the inputs to the RRBFNNs.


 (µ11 , · · · , µn 1 , . . . , µ12 , . . . , µn 2 , · · · ,  3: Presume that Wk∗ matrices are limited
 
i i
Assumption

∗

 
as Wk F ≤ Wk,M , where k·kF indicates the Frobenius
 µ1nj , · · · , µni nj ),
 



norm [51], [52].
 (σ11 , · · · , σn 1 , . . . , σ12 , . . . , σn 2 , · · · ,
 

 i i  It is noticed that the restricted values Wk,M are not needed
σ1nj , · · · , σni nj ). to carry out the designed control system. However, those
values are still necessary to perform the stability analysis. The
C. CONVERGENCE ANALYSES weights of the RRBFNN uncertainty observer can be trained
through applying the Taylor series expansion of ∗k (xk Wk∗ )

In this paper, the convergence analyses are introduced to
derive the learning rates parameters to assure convergence around Ŵk . Thus, the error function can be optimized

of the output error of RRBFNN parameters. Therefore, as [51]–[57]:
the learning rates parameters selection has a significant effect 
∗k (xk Wk∗ ) − 

on the RRBFNN performance. In order to train the RRBFNN ˆ k (xk Ŵk )
effectively, six varied learning rates, ηW , ηo ηµ , ησ , ηβ and ηα ,
which guarantee the convergence of tracking errors and iden- = W̃kT 4k + Hk (Wk∗ , Ŵk ) (30)
" #T
tification based on the analyses of a discrete-type Lyapunov ∂ ˆ k,1 ∂ 
ˆ k,2 ∂ ˆ k,n
function, are derived in the Appendix [27], [49], [50]. 4k = , ,··· , , (31)
∂ Ŵk,1 ∂ Ŵk,2 ∂ Ŵk,n
D. THE PROPOSED RABC SYSTEM DESIGN where k = r, qs, ds and the vectors with higher order terms
The main goal of this paper is to design a robust and effi- are represented by W̃k = (Wk∗ − Ŵk ) and Hk (Wk∗ , Ŵk ).
cient control method for the HSPMSM model described by Substituting (30) into (31) will provide:
(9)-(15) regardless load disturbances and obscure parame- 
ter changes. Accordingly, an RABC scheme employing the
RRBFNN-based uncertainty observer was designed as shown
in Fig. 1.Thus, the rotor position state trajectory θr (t) can
asymptotically follow the reference state trajectory position
θr∗ (t).We presume that θr∗ (t) and θ̇r∗ (t) are the time functions
with additional constraints.
To accurately detect the lumped parameter uncertainties
k (xk |Wk ) = ˆ k (xk Ŵk ) + W̃kT 4k + γk − uRC
k (32)
where γk = Hk (Wk∗ , Ŵk ) + εk . The uncertainty expression γk
is presumed to be limited by kγk k ≤ σk while σk are positive
constants. Due to the uncertainty observer errors (tracking,
estimation, and weight), we can use σk to represent the flaw
of the proposed RRBFNN design. Rewrite (32) will provide:
described in (13)-(15), the RRBFNN-based observer out- 
k (xk |Wk ) = 
ˆ k (xk Ŵk ) + W̃kT 4k + γk − uRC (33)
puts  ˆ r (·), 
ˆ qs (xqs ) and 
ˆ ds (xds ) are derived according k
to assumption 1 and the universal approximation theorem The design procedures of the overall proposed RABC sys-
in [51], [52] as: tem including the RRBFNN uncertainty observer are summa-
k (xk ) = Wk∗T 8∗k + εk (27) rized as follows:
Procedure 1: Determine the state function of the tracking
where Wk∗ and 8∗k represent the optimal parameter matrices of error:
the proposed observer while k = r, qs, ds; εk is the minimum
remodeled error. Although these optimal parameter matrices z1 (t) = θr (t) − θrm (t) (34)
cannot be obtained, although, it can be estimated using the
Thereafter, by differentiating the error function as:
RRBFNN uncertainty observer as:
ż1 (t) = θ̇r (t) − θ̇rm (t) = ωr − θ̇rm (t) (35)

ˆ k (xk ) = ŴkT 8̂k + uRC
k (28)
where θ̇r (t) = ωr (t) can be noted as virtual control in (35).
where Ŵk and 8̂k are the estimated values of the optimal
A Lyapunov function nominee can be adopted as:
parameter matrices Wk∗ and 8∗k , respectively; uRC k expresses
the robust controller function that recover the flaw of the pro- 1 2
V1 (t) = z (t) (36)
posed RRBFNN uncertainty observer as a result of tracking, 2 1
VOLUME 7, 2019 99333
 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

By differentiating (36) then substituting from (35) as: Moreover, the derivative of (46) will introduce:
V̇1 (t) = z1 ż1 = z1 [θ̇r (t) − θ̇rm (t)] (37) V̇3 (t) = V̇2 (t) + z3 (t)ż3 (t)
 
Subsequently, set the virtual control law x̄2 (t) = ωr∗ as = −k2 z22 +z3 (t) f2 +(Vqsr
/Lss ) + qs − i̇r∗
qs (t) (47)
follows: The q-axis desired control law Vqs r∗ is prepared as:

x̄2 = −k1 z1 (t) + θ̇rm (t) (38)

 
r∗
Vqs = Lss −k3 z3 −f2 − ˆ qs + i̇r∗
qs (t)
where k1 > 0 is defined as again value for the control  
= Lss −k3 z3 −f2 − Ŵqs 4qs − uRC
T r∗
qs −σqs + i̇qs (t) (48)
design.
Procedure 2: Repeat procedure 1 through defining the where k3 > 0 is defined as a design control gain; the q-axis
tracking error state function of the rotor speed: estimated value of the approximated parameter uncertainty
qs of the RRBFNN-based observer is defined as  ˆ qs =
z2 (t) = x2 (t) − ωr∗ (t) (39) Ŵqs 4qs + uqs + σqs .Thus, the state function of the tracking
T RC

Therefore, by utilizing (10) and (27)-(29), the error error can be determined as:
function is differentiated as: z4 (t) = x4 (t) − ir∗
ds (t) (49)
ż2 (t) = ẋ2 − ω̇r∗ (t) Procedure 4: By using the derivative of (49) then substitut-
ing from (12) yields:
= −(βm /Jm )x2 + (Kt /Jm )x3 + r − ω̇r∗ (t)
ż4 (t) = ẋ4 (t) − i̇r∗
ds (t)
= f1 + (Kt /Jm )x3 + r − ω̇r∗ (t) (40) r
V
where f1 = −(βm /Jm )x2 . = f3 + ds + ds − i̇r∗ ds (t) (50)
Lss
The candidate of the Lyapunov function is selected as:
where f3 = −(Rs /Lss )x4 + x2 x3 .
1 Subsequently, select the candidate of the Lyapunov
V2 (t) = V1 (t) + z22 (t) (41)
2 function as:
By differentiating V2 (t) then utilizing (39) and (40): 1
V4 (t) = V3 (t) + z24 (t) (51)
2
V̇2 (t) = V̇1 (t)+z2 (t)ż2 (t)
Thereafter, (51) is differentiated and substituting from
βm
 
Kt
= −k1 z1 + z2 (t) − x2 + x3 +r − ω̇r (t)
2 ∗
(42) (47) as:
Jm Jm
V̇4 (t) = V̇3 (t) + z4 (t)ż4 (t)
The virtual control law x̄3 (t) = ir∗
qs is constructed as:

Vr

  = −k3 z23 + z4 (t) f3 + ds + ds − i̇r∗
ds (t) (52)
x̄3 = (Jm /Kt ) −k2 z2 − f1 − ˆ r + ω̇r∗ (t) Lss
The d-axis desired control law Vds r∗ is derived as:
 
= (Jm /Kt ) −k2 z2 −f1 − ŴrT 4r −uRC r −σr + ω̇r
∗
(t) (43)  
r∗
Vds = Lss −k4 z4 −f3 −  ˆ ds + i̇r∗
ds (t)
where k2 > 0 is defined as a design control gain; the  
estimated value of the approximated parameter uncertainty = Lss −k4 z4 − f3 − ŴdsT
4ds − uRC e∗
ds −σds + i̇ds (t) (53)
r of the RRBFNN-based observer is defined as  ˆr =
where k4 > 0 is a defined as a design control gain; the d-axis
ŴrT 4r + uRC
r + σr . Thus, the tracking error state of the rotor
estimated value of the approximated parameter uncertainty
speed can be defined as:
ds of the RRBFNN-based observer is defined as  ˆ ds =
T 4 + uRC + σ .
Ŵds
z3 (t) = x3 (t) − x̄3 (t) (44) ds ds ds

Procedure 3: The derivative of (44) and employing (11) as: E. STABILITY ANALYSIS OF THE PROPOSED RABC SYSTEM
This section aims to examine the stability analysis of the
ż3 (t) = ẋ3 (t) − x̄˙3 (t) proposed RABC process with a RRBFNN-based uncertainty
r
Rs λ0 Vqs observer for the motor drive system operation. The candidate
=− x3 − x2 x4 − m x2 + + qs − i̇r∗
qs (t) of the Lyapunov function can be considered as follows:
Lss Lss Lss
r
Vqs 3
+ qs − i̇r∗ 1X T 1 1
= f2 + qs (t) (45) Va = zυ zυ + tr(W̃rT 0r−1 W̃rT )+ tr(W̃qs 0qs W̃qs )
T −1 T
Lss 2 2ηr 2ηqs
υ=1
where f2 = (Lss )−1 −Rs x3 − x2 x3 Lss − λ0m x2 .


1 1 −1 2 1 −1 2
Select the candidate of the Lyapunov function to be: + 0ds W̃ds )+
T −1 T
tr(W̃ds 0 σ̃ + 0 σ̃
2ηds 2ηr r r 2ηqs qs qs
1 1 −1 2
V3 (t) = V2 (t) + z23 (t) (46) + 0 σ̃ (54)
2 2ηds ds ds

99334 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

where 0qs = diag[0qs,i ], 0qs = diag[0qs,i ], and 0ds = = −k1 z21 − k2 z22 − k3 z23
diag[0ds,i ], i = [1, 2, · · · , n], and 0r,i , 0qs,i , and 0ds,i are
r + σr ) + z2 (−uqs + σqs )
+z1 (−uRC RC
the tuning gains. tr(·) indicates the trace of a matrix.
ds + σds )+z1 W̃r 4r +z2 W̃qs 4r +z3 W̃ds 4ds
+z3 (−uRC T T T
Assumption 4: Suppose that ξ is positive constant and the
Lyapunov function candidate denoted in (54) is limited as 1 ˙ ) − 1 tr(W̃ T 0 −1 Ŵ˙ )
− tr(W̃rT 0r−1 Ŵ r qs
Va ≤ ξ . ηr ηqs qs qs

Stability Theorem: The HSPMSM drive system is demon- 1 T −1 ˙ 1

− tr(W̃ds 0ds Ŵds ) − 0r−1 σ̃r σ̂˙ r
strated by (9)-(15) while uncertainties are considered. The ηds ηr
adaptive control laws, represented by (38), (43), (48) and 1 −1 ˙ 1 −1 ˙
(53), are used to design the proposed RABC with a RRBFNN- − 0qs σ̃qs σ̂qs − 0 σ̃ds σ̂ds (68)
ηqs ηds ds
based uncertainty observer. Furthermore, if the control sys-
tem fulfills Assumptions (1-4) and the adaptation control Compensating (55)-(63) into (68) will provide:
laws are selected as (55)-(57) for all weights of the RRBFNN V̇a = −k1 z21 − k2 z22 − k3 z23 − k4 z24
observer, then the stability of the designed RABC can be
r +σr )+z3 (−uqs + σqs )+z4 (−uds + σds )
+z2 (−uRC RC RC
ensured via designing the robust controllers as (58)-(60) with
the adaptive estimations to be limited as (61)-(63). 1 1 −1 ˙ 1 −1 ˙
− 0r−1 σ̃r σ̂˙ r − 0 σ̃qs σ̂qs − 0 σ̃ds σ̂ds
˙ = 0 4 z − η 0 Ŵ ηr ηqs qs ηds ds
Ŵ r,i r,i r,i 2,i r r,i (55)
r,i
˙ ≤ −k1 z21 − k2 z22 − k3 z23 − k4 z24
Ŵ qs,i = 0qs,i 4qs,i z3,i − ηqs 0qs,i Ŵqs,i (56)
r − |σr |] − |z2 | [uqs − σqs ]
− |z1 | [uRC RC
 
˙
Ŵds,i = 0ds,i 4ds,i z4,i − ηds 0ds,i Ŵds,i (57)
 
− |z3 | [uRC
ds − |σds |]
r,i = σ̂r,i sgn(z2,i )
uRC (58)
uqs,i = σ̂qs,i sgn(z3,i )
RC
(59) ≤ −k1 z21 − k2 z22 − k3 z23 − k4 z24 ≤ 0 (69)
uRC
ds,i = σ̂ds,i sgn(z4,i ) (60) Since V̇a (Z (t), W̃k , σ̃k (t)) is a negative semidefinite func-
σ̂˙ r,i = ηr (z2,i ) tion (i.e. Va (Z (t), W̃k , σ̃k (t)) ≤ Va (Z (0), W̃k , σ̃k (0)), which
 
(61)
σ̂˙ qs,i = ηqs (z3,i )
 
(62) denotes that Z (t), W̃k and σ̃k (t) are limited functions.
Determine the subsequent term as:
σ̂˙ ds,i = ηds (z4,i )
 
(63)
2a (t) ≡ k1 z21 + k2 z22 + k3 z23 + k4 z24 ≤ −V̇a (Z (t), W̃k , σ̃k (t))
where ηr , ηqs and ηds are positive constants; 4r,i , 4qs,i and
4ds,i are the ith elements of 4r , 4qs and 4ds , respectively. (70)
σr , σqs and σds represent the terms of uncertainty which can where Z = [z1 , z2 , z3 , z4 ], k = r, qs, ds. Therefore:
be estimated online by σ̂r , σ̂qs and σ̂ds , respectively. The
sgn(·) indicates the sign function. Eventually, upon satisfy- Zt
ing Assumption 4 with any initial conditions, the adjustable 2a (τ )dτ ≤Va (Z (0), W̃k , σ̃k (0)) − Va (Z (t), W̃k , σ̃k (t))
weights Ŵ ˙ , Ŵ ˙ ˙
r,i qs,i and Ŵds,i and the errors of states Z = 0
[z1 , z2 , z3 , z4 ] of the closed loop system are symmetrically (71)
restricted and can be maintained at arbitrary small value.
Since Va (Z (0), W̃k (0), σ̃k (0) is a limited function while
Proof of Stability Theorem:
Va (Z (t), W̃k (t), σ̃k (t)) is a limited and non-rising function,
The tracking error states can be differentiated as Ż =
the next outcome can be acquired as:
[ż1 , ż2 , ż3 , ż4 ] by substituting (38), (43), (48) and (53)
into(35), (40), (45) and (50), respectively, as follows: Zt
ż1 (t) = −k1 z1 (64) lim 2a (τ )dτ ≤ ∞ (72)
t→∞
ż2 (t) = −k2 z2 + W̃rT 4r + σr − uRC
r (65) 0
T
ż3 (t) = −k3 z3 + W̃qs 4r + σqs − uRC
qs (66) Since 2̇a (t) is a limited function, consequently 2a (t) is a
T
ż4 (t) = −k4 z4 + W̃ds 4ds + σds − uRC
ds (67) symmetrical continuous function. By employing Barbalat’s
Lemma [52], it can be demonstrated that:
If we differentiate the candidate of the Lyapunov function
(54) and utilizing (64)-(67), we will attain the following: lim 2a (t) = 0 (73)
t→∞
V̇a = zT1 ż1 + zT2 ż2 + zT3 ż3 + zT4 ż4
We can notice that as t → ∞, the function Z (t) will converge
1 ˙ ) − 1 tr(W̃ T 0 −1 Ŵ
˙ )
− tr(W̃rT 0r−1 Ŵ r qs to zero. As a result, the stability of proposed RABC with
ηr ηqs qs qs
RRBFNN-based uncertainty observer is guaranteed.
1 T −1 ˙ 1
− tr(W̃ds 0ds Ŵds ) − 0r−1 σ̃r σ̂˙ r Remark 2: The lumped uncertainty terms σr , σqs and σds
ηds ηr comprise optimal parameters of the network, approximation
1 −1 ˙ 1 −1 ˙ errors, and higher order terms of Taylor series. Accordingly,
− 0qs σ̃qs σ̂qs − 0 σ̃ds σ̂ds
ηqs ηds ds a conservative control law with considerable limits is selected

VOLUME 7, 2019 99335

 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

due to the unavailability of those terms in practical appli-

cations. In addition, the selection process of the upper limit
of the uncertainty terms σr , σqs and σds has a considerable
influence on the performance of the control system. If the
limits are chosen too large, the sign function of the controller
may lead to a significant chattering incident in the control
attempts. The unwanted chattering control attempts will trig-
ger unsteady system dynamics. Contrarily, if the limits are
chosen too small, the stability conditions may not be fulfilled.
Thus, the HSPMSM drive system will be unstable. Hence,
the adaptive bound estimation algorithms in (61)-(63) are
used in this paper o simplify the adjustment of the limits
in real time for the HSPMSM drive system based RRBFNN
uncertainty observer.
Remark 3: According to Remark 1 and by comparison
the RABC scheme with the backstepping control techniques
in [19]–[26], the proposed strategy can resolve the problem of
performance degradation by assessing the uncertainty terms
in the dynamic model (9)-(15), r , qs and ds . In addition,
the model uncertainties qs and ds of the drive system were
not considered in [19]–[26]. Though, if the HSPMSM drive
system parameters are perturbed (13)-(15), the control laws
in [19]–[26] which consider only the mechanical uncertainty
term (r ) will lead to unstable drive system and the system
performance may be degraded because the uncertainty terms
qs and ds are not considered in the design step of the
control method. The control laws in [19]–[26] only contains
the neural network output. However, the control laws (43),
(48) and (53) in this paper not only contains the RRBFNN
output, which is utilized to estimate the parameter uncer-
FIGURE 3. The schematic diagram of the overall developed DSP-based
tainties, but also the developed robust controllers (58)-(60), high-speed drive system.
which are intended to conform the values of lumped
parameter uncertainties using the adaptive laws (61)-(63).
Furthermore, the learning algorithms were only utilized to is shown in Fig. 3. To implement the control operation,
adjust the thresholds and weights of the neural network in a DSP-based development controller board (dSPACE
[19]–[26] so as to give proper control performance. Nonethe- DS1102) with a TMS320C31 and TMS320P14 digital signal
less, in this paper, the learning algorithms (55)-(57), are processors is utilized. The control board involves several
utilized to online adapt the interior feedback, the center input/output ports (PIO, ADC, DAC, and encoder) to acquire
parameters, and the width parameters. Hence, to assure the the measured signals and send the proper control actions.
the HSPMSM drive system stability in spite of the extrinsic To enhance the precision of the measured feedback signals
load disturbance and dynamics of parameter uncertainties (position and speed), the encoder interface circuits uses a
existed in (13)-(15), the RABC-based RRBFNN observer is digital filter with frequency multiplied by four. The PWM
suggested to compensate all these parameter uncertainties. signals of the inverter are generated based on a carrier fre-
quency of 15 kHz which provides a sampling rate of 66.67µs.
IV. VALIDATION RESULTS The position control loop utilizes a time interval of 1 ms.
In this section, the simulation and experimentation tests A six-IGBT switches were used to build the current-regulated
are performed to investigate the effectiveness of the pro- PWM VSI. A 10000 pulses/revolution incremental optical
posed RABC scheme. The simulation tests are imple- encoder was applied to carry out the position acquisition.
mented through MATLAB/SIMULINK software according Consequently, a high precision measurement of the position/
to the control schemes demonstrated in Figs. (1, 2). The speed is resulted due to the high output frequency of the
schematic diagram of the experimental hardware setup is multiplier circuit (40000 pulses/revolution). Furthermore,
shown in Fig. 3. the computed torque controller (CTC), the IBC and the pro-
posed RABC schemes are implemented. Figure 4 shows the
A. EXPERIMENTAL SET-UP software flowcharts of the proposed RABC using RRBFNN.
The block diagram of the proposed control scheme with The proposed real-time control algorithm implementa-
DSP-based controller for the high speed PMSM drive system tion process consists of the main control program along

99336 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design
FIGURE 4. Flowcharts of the RABC algorithm.
with its subroutines. First, the initialization process of the
input/output (I/O) and system parameters is set. After,
the intervals for the two interrupt routines (IR1 and IR2) are
set. Later, the counters of the encoder circuits are initialized
by setting the servo drive. Once the interrupt is enabled,
the main program is applied to observe the data of the control
system. A sampling period of 1 ms is chosen to accomplish
CPU calculations with high performance of the proposed
RABC algorithm. The first interrupt subroutine (IR1) is uti-
lized for the implementation of the control algorithms and
the interface operation of the encoders. Initially, IR1 uses
the encoders to examine the position of the HSPMSM. Next,
IR1 with 1 ms sampling rate is utilized to calculate tracking
error states (z1 , z2 , z3 and z4 ) and its derivatives (ż1 , ż2 , ż3
and ż4 ), the virtual control laws x̄2 and x̄3 , real-time training
of the RRBFNN and computation of the parameter uncertain-
ties from the RRBFNN observer, calculation of the robust
controllers, the adaptive control laws computation, the esti-
mation of the adaptive limit algorithm, calculation of the
VOLUME 7, 2019
RABC algorithm and updating the weights of the RRBFNN.
Later, the IR2 with 0.2 ms sampling rate is utilized to col-
lect the encoder data, perform the abc/d-q transformation,
determine the d-q command currents, and perform d-q axis
reference SVPWM voltages to generate the switching signals
that control the inverter operation. Considering the stability
needs and different operating condition, the parameters of
the proposed RABC scheme are selected to accomplish the
preferable tracking performance. An online parameter learn-
ing technique is used to retrieve the inaccurate initialization
of system parameters. Hence, the adjustment operation of
system parameters is regularly active for the whole running
duration of experiments. Furthermore, the proposed RABC
scheme parameters are: ηds = ηqs = 6.0, ηr = 3.0, k1 = 9.5,
k2 = 3.5, k3 = 7.5 and k4 = 7.5.
B. SIMULATION RESULTS
To investigate the feasibility of the proposed RABC scheme,
the HSPMSM servo drive system is simulated and tested
under various operating conditions. Subsequently, four dif-
ferent operating conditions of extrinsic load perturbations
and parameter uncertainties (PUs) are studied to examine the
robustness of the proposed controllers as follows:
Case 1: 1.0 × (Ls /Rs ), 1.0 × (βm /Jm ), 1.00 × λm ,
TL = 0 − 0.5 mN.m
Case 2: 0.5 × (Ls /Rs ), 1.5 × (βm /Jm ), 0.85 × λm ,
TL = 0 − 0.5 mN.m
Case 3: 1.5 × (Ls /Rs ), 2.5 × (βm /Jm ), 1.25 × λm ,
TL = 0 − 0.5 mN.m
Case 4: 1.5 × (Ls /Rs ), 5.0 × (βm /Jm ), 1.25 × λm ,
TL = 0 − 0.5 mN.m
For Case 1, we investigate the dynamic performance of
the HSPMSM servo drive system under external loading
command change 0-0.5 mN.m for the both IBC and RABC
schemes while parameter ratios are maintained constant.
Fig. 5 (a) shows the dynamic performance of the drive system
with IBC scheme in terms of the reference and actual rotor
positions, the position tracking error, the reference and actual
rotor speeds, the speed tracking error, and d −q axis currents,
respectively. Moreover, the dynamic performance using the
RABC scheme is investigated at the same operating condi-
tions as depicted in Fig. 5 (b). At t = 2.5sec, the motor shaft is
loaded by 0.5 mN.m then load is removed after at t = 7.5sec
in order to check the disturbance rejection capabilities of
both IBC and the proposed RABC. As seen from Fig. 4,
the simulation results achieve good dynamic performances
for both controllers under command change and load reg-
ulation. We can observe that the proposed RABC scheme
provides better performance then IBC in load regulation and
command tracking characteristics. Consequently, the results
obtained from Fig. 4 show larger position and speed tracking
errors with IBC scheme compared to the proposed RABC
scheme with RRBFNN-based uncertainty observer. Further-
more, the results have demonstrated a substantial reduction
of the utmost dip of both rotor position and speed with the
proposed RABC scheme.
99337
 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

FIGURE 5. Simulation results of the dynamic performance of the HSPMSM servo drive system with a reference position model of
π rad and subsequent loading of 0.5 mN.m using: (a) the IBC and (b) the proposed RABC with RRBFNN-based uncertainty observer.

99338 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

FIGURE 6. Simulation results for all different Cases (1-4) of the enlarged dynamic performance under PUs of the HSPMSM servo drive system with a
reference position model of 2π rad and subsequent loading of 0.5 mN.m using: (a) the IBC and (b) the proposed RABC with RRBFNN-based uncertainty
observer.

In addition, the external load disturbance and PUs are capability of the proposed RABC scheme. Fig. 6 and 7 show
further detailed investigated through four different cases the comparative dynamic performance at all Cases of PUs
(Case 1-4) to be compared in order to confirm the robustness of the HSPMSM servo drive system for both the IBC and

VOLUME 7, 2019 99339

 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

FIGURE 7. Simulation results for all different Cases (1-4) of the load regulation characteristics under PUs of the HSPMSM servo drive system using:
(a) the IBC and (b) the proposed RABC with RRBFNN-based uncertainty observer.

the proposed RABC schemes. It is clearly noticed that the under the incident of PUs. Thus, the tracking errors have been
tracking errors quickly converge to zero which validate the significantly reduced as well as load regulation capabilities
robustness characteristics of the proposed RABC scheme have been verified compared to the IBC scheme. As a result,

99340 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

FIGURE 8. Experimental results of the dynamic performance of the HSPMSM servo drive system with a reference position
model of 2π rad and subsequent loading of 0.5 mN.m using: (a) the IBC and (b) the proposed RABC with RRBFNN-based
uncertainty observer.

VOLUME 7, 2019 99341

 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design
FIGURE 9. Experimental results of the dynamic performance of the HSPMSM servo drive system with a reference position model
of 2π rad at no-load condition using: (a) IBC and (b) proposed RABC with RRBFNN-based uncertainty observer.
the designed RABC structure can provide preferable control
response compared to the IBC structure. In addition, the pro-
posed RABC scheme yields a faster response (within 0.2 sec)
and higher precision than the IBC scheme for the reference
model under load variations. On the other hand, the IBC under
PUs has indolent recovery time (>2.5 sec). Eventually, for all
cases of PUs, it can be confirmed that the proposed RABC
scheme provides several advantages in terms of its tracking
accuracy, robustness, as well as suitability with the HSPMSM
control system
C. EXPERIMENTAL RESULTS
A hardware experimental prototype of the HSPMSM with the
same simulation parameters was tested to validate the high
performance of the developed RABC scheme compared to
the IBC scheme. The laboratory tests were performed based
on the control schemes presented in Figs. (1-4).
To further investigate the effectiveness of the developed
control schemes for micro drive-based industrial applications,
experimental test findings are presented. Fig. 8 shows com-
parative test results for the dynamic performance of the IBC
versus the developed RABC under desired model command
with subsequent applied loading condition of 0.5 mN.m. The
dynamic responses of the IBC involving the reference and
actual rotor positions, the position tracking error, the ref-
erence and actual rotor speeds, the speed tracking error,
the d − q axis currents, and adaptive position/speed sig-
nals are depicted in Fig. 8(a). Furthermore, the dynamic
responses of the proposed RABC including same signals are
99342
shown in Fig. 8(b). In addition, the results have shown better
disturbance rejection capability with the proposed RABC
scheme compared to the IBC scheme. At full load condi-
tion, we can also notice that the proposed RABC with the
RRBFNN-based uncertainty observer provides less maxi-
mum position tracking errors of ∼0.15 rad while the IBC has
higher error of ∼0.6 rad. Obviously, the test results acquired
in Fig. 8 show better dynamic response of the proposed
RABC scheme accomplishing preferable load regulation and
command tracking. Moreover, the drive system performance
has been investigated for both IBC and RABC schemes
at no-load condition as displayed in Fig. 9. It is clearly
illustrated that the proposed RABC gives better regulation
characteristics as well as significantly reduced position/speed
tracking errors compared with the IBC. Subsequently, the
proposed RABC scheme validates its superiority perfor-
mance compared with the IBC. Undoubtedly, the RABC
scheme with RRBFNN-based uncertainty observer has veri-
fied its preferable performance for the HSPMSM drive appli-
cations with greatly improved characteristics to a great extent.
Consequently, it was proved that the developed RABC design
accomplishes the precision demands, robustness, and suit-
ability for high performance industrial drive applications.
D. EVALUATION AND COMPARISON OF
CONTROL PERFORMANCE
To evaluate the performance of the servo drive system,
we will use three tracking error indices (maximum, average,
and standard deviation) [18], [40]. The control response can
VOLUME 7, 2019
F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

be readily compared utilizing (74)-(76) as follows:

p
TEmax = max T (k)2 (74)
k
n
X T (k)
TEmean = (75)
n
k=1
v
u n
uX (T (k) − Tmean )2
TEsd =t (76)
n
k=1

where TEmax is the maximum tracking error, TEmean is the

average tracking error, TEsd is the standard deviation of the
tracking error, T (k) = [θrm (k) − θr (k)].

FIGURE 11. Tracking errors reduction percentage using IBC and proposed
RABC schemes for HSPMSM servo drive system (experimentation):
(a) TEmax, (b) TEmean, and (c) TEsd.

TABLE 2. Performance evaluation of the HSPMSM servo drive system

(experimentation).

the maximum, average, standard deviation tracking errors

FIGURE 10. Performance measures of CTC, IBC and proposed RABC
schemes for HSPMSM servo drive system (experimentation): (a) TEmax,
(b) TEmean, and (c) TEsd.
The performance evaluation of the different position con-
trollers schemes are depicted in Fig. 10. The various compara-
tive performance average measures with respect to computed
torque controller (CTC) show that the proposed RABC pro-
vides lower tracking errors by: 87.66% for maximum error,
96.48% for average error and 94.74% for standard deviation
error. In regard to IBC compared to CTC, the averages of
are decreased by 56.36%, 75.10% and 81.13%, respectively.
The percentage reductions of the tracking errors utilizing the
IBC and the proposed RABC schemes compared to the CTC
scheme are shown in Fig. 11. The comparative analysis of
the control performance is illustrated in Table 2. Furthermore,
Table 3 depicts the improvement in the tracking errors with
IBC and proposed RABC in comparison to the CTC. It is
apparent that the performance measures of the HSPMSM
servo drive system are significantly enhanced with the pro-
posed RABC scheme. Consequently, the developed RABC
with RRBFNN-based uncertainty observer fulfills the high
precision demands. Thus, the proposed scheme has verified

VOLUME 7, 2019 99343

 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

TABLE 3. Percentage reductionof tracking errors based on CTC and the actual state; n = 1, 2, · · · , 4. The learning algorithm
(experimentation).
based on the BP is described as follows.
Layer 3: In the output layer, the error term to be propagated
is calculated as:
∂En
δo3 (N + 1) = − (N + 1)
∂neto3
∂En ∂znn ∂En ∂znn ∂u
   
= − n 3 = − n (A.2)
its superiority for the position/speed control of the HSPMSM ∂zn ∂yo ∂zn ∂u ∂y3o
servo drive systems for industrial applications. The weight is updated by the amount:

V. CONCLUSION 1Wj (N + 1)
In this paper, we proposed an RABC using RRBFNN uncer- ∂En
tainty observer for HSPMSM adjustable speed drives to = −ηW (N + 1)
∂Wj
achieve high dynamic control performance in the existence of
∂En ∂y3o ∂neto3
 
parameter uncertainties and extrinsic load perturbations. The = −ηW 3 = ηW δo3 (N + 1)82j (A.3)
RABC scheme encompasses an IBC, a RRBFNN uncertainty ∂yo ∂neto3 ∂Wj
observer and a robust controller. First, the IBC is designed 1Wo (N + 1)
in the sense of Lyapunov stability theorem to stabilize the ∂En
HSPMSM drive system and to satisfy multiple objectives = −ηo (N + 1)
∂Wo
of a stable rotor position to trace the desired trajectory.
∂En ∂y3o ∂neto3
 
Though, certain information about the lumped parameter = −ηo 3 = ηo δo3 (N + 1)ylo (A.4)
uncertainties are needed within the backstepping control law ∂yo ∂neto3 ∂Wo
so that the system performance would not severely affected. 1βol (N + 1)
Consequently, an RABC was developed to enhance the ∂En
robustness of the HSPMSM drive system as a result of extrin- = −ηβ l (N + 1)
∂βo
sic load perturbations as well as parameter uncertainties.
∂En ∂y3o ∂neto3
 
Thus, an online adaptive observer based on RRBFNN was = −ηo 3 = ηβ δo3 (N + 1)ylo (A.5)
incorporated to estimate the nonlinear parameter uncertain- ∂yo ∂neto3 ∂βol
ties. Moreover, the robust controller was developed to retrieve where ηW , ηo and ηβ are the learning rate parameter of the
the remaining of the RRBFNN estimate errors. To assure the connecting weights between the hidden layer and the output
stability of the developed RABC, the Lyapunov stability anal- layer of the RRBFN. The weights of the output layer are
ysis has been used to derive the online adaptive control laws. updated according to the following equations.
Experimental tests of the HSPMSM drive system were per-
formed to confirm the validation of the designed RABC Wj (N + 1) = Wj (N ) + 1Wj (N + 1) (A.6)
scheme. The dynamic performance of the HSPMSM drive Wo (N + 1) = Wo (N ) + 1Wo (N + 1) (A.7)
system has been studied under wide range of operating
conditions. The test results assure an improved dynamic βol (N + 1) = βol (N ) + 1βol (N + 1) (A.8)
response and robust control performance of the developed Layer 2: In the hidden layer, the error term is calculated as:
RABC regardless the parameter changes and load distur-
3 ∂net 3 ∂y2
!
bances. In conclusion, the main contributions of this paper ∂E n ∂E n ∂E n ∂y j
can be summarized as: a novel RABC was successfully devel- δj2 = − =− j = − 3 o o
∂netj2 ∂ Ãi ∂yo ∂neto3 ∂y2j ∂ Ãj
oped, implemented, and applied for the HSPMSM drive sys- i
X
tem to achieve robust control performance considering load = Wj 8j (A.9)
disturbances and parameters uncertainties; a new model of j
RRBFNN-based online uncertainty observer was effectively j j
designed to conform the nonlinear uncertainties. The update laws of the µi and the σi are given by:
j
" #
j ∂En ∂En ∂∂ Ãi
APPENDIX 1µi (N + 1) = −ηµ j (N + 1) = −ηµ j j
A. ON-LINE LEARNING ALGORITHM OF THE RRBFNN ∂µi ∂∂ Ãi ∂µi
To describe the on-line parameter learning algorithm, first the j
= ηµ δj2 (N + 1) · Ãi (8i )
j
energy function En is defined as  
· (8j − µj )T σj (8j − µj ) (N − 1)
En = (1/2)(xnd − xn )2 = (1/2)(zn )2 (A.1)
j
!
where xnd (t) is the reference command, xn (t) is the actual state ∂8i
× j
(N ) − 1 (A.10)
and zn is the error signal between the reference command ∂µi

99344 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

j
∂8i j j B. CONVERGENCE ANALYSES OF THE RRBFNN
j
(N ) = ζi (N − 1)αi (N )
∂µi Consider the energy function in (A.1) as a discrete-type
  Lyapunov function and the change in the Lyapunov function
· (8j − µj )T σj (8j − µj ) (N − 1) can be written as:
j
!
∂8i
× j
(N ) − 1 (A.11) En (N + 1)
∂µi
" j
# = En (N ) + 1En (N )
j ∂En ∂En ∂ Ãi ny no 
1σi (N + 1) = −ησ j
(N + 1) = −ησ j j
X X ∂En (N ) 
∂σi ∂ Ãi ∂σi ≈ En (N ) + 1Wj
j j
∂Wj
= ησ δj2 (N + 1) · Ãi (8i ) j=1 o=1
ny no  ny no 
  X ∂En (N )  X X ∂En (N ) 
· (8j −µj )T σj (8j −µj ) (N −1)
X
(A.12) + 1Wo + 1βol
∂Wo ∂βol
j l=1 o=1 l=1 o=1
∂8i j j nj "
= ζi (N −1)αi (N ) ni X
#
j
(N ) X ∂En (N ) j ∂En (N ) j ∂En (N ) j
∂σi + j
1µi + j
1σi + j
1αi
 
i=1 j=1 ∂µi ∂σi ∂αi
· (8j −µj )T σj (8j −µj ) (N −1) (A.13)
(B.1)
j
The update weight of the feedback, αi , is: En (N + 1)
ny no 
j 1 X ∂En (N ) ∂y3 2
1αi (N + 1)
X
= En (N ) − ηW o
" j
# 6 ∂y3o ∂Wj
∂En ∂En ∂ Ãi j=1 o=1
= −ηα (N + 1) = −ηα ny no
∂αi
j j
∂ Ãi ∂αi
j
1 X X  ∂En (N ) ∂y3 2
+ En (N ) − ηo o
= ηα δj2 (N + 1) · Ãi (8i )
j j 6 ∂y3o ∂Wo
l=1 o=1
ny no 
j X ∂En (N ) ∂y3 2
!
  ∂8i 1 X
· (8j − µj ) σj (8j − µj ) (N )
T
(N ) (A.14) + En (N ) − ηβ o
j
∂αi 6 ∂y3o ∂βol
l=1 o=1
j
∂8i ni Xnj no !2
(N ) 1 X X ∂En (N ) ∂xni ∂y3o
j + En (N ) − ηµ
∂αi 6 ∂xni ∂y3o ∂µj
i=1 j=1 o=1 i
j j
= ζi (N − 1)αi (N ) ni Xnj no !2
1 X X ∂En (N ) ∂xni ∂y3o
j + En (N ) − ησ
!
∂8i
∂xni ∂y3o ∂σ j
 
· (8j − µj )T σj (8j − µj ) (N − 1) j
(N ) − 1 6
i=1 j=1 o=1 i
∂αi !2
ni Xnj no
j 1 X X ∂En (N ) ∂xni ∂y3o
+ζi (N − 1) (A.15) + En (N ) − ηα (B.2)
6 ∂xni ∂y3o ∂αi
j
where ηµ and ησ and ηα are the learning rate parameters i=1 j=1 o=1
of the mean, standard deviation and the self-feedback loop,
respectively. Moreover, they can be updated as follows: where 1Wj , 1Wo , 1µij , 1σji , 1βol and 1αji represent the
µij (N + 1) = µij (N ) + 1µij (N + 1) (A.16) weight change in the output layer, the mean and standard
deviation in the Gaussian function and the weight change
σji (N + 1) = σji (N ) + 1σji (N + 1) (A.17) in the self-feedback loops, respectively. If the learning rate
αji (N + 1) = αji (N ) + 1αji (N + 1) (A.18) parameters of the RRBFNN are designated as
The exact calculation of the Jacobian of the HSPMSM drive
En (N )
system (∂xn /∂y3o ) in (A.2), cannot be easily determined and ηW = " # (B.3)
ny P
no 2
need heavy computation due to the uncertainties of the servo 
∂En (N ) ∂y3o
+ν
P
6 ∂yo ∂Wj
3
drive system dynamic, such as parameter variations and j=1 o=1
external load disturbances. To overcome this problem and to En (N )
increase the online learning speed of the network parameters, ηo = " # (B.4)
ny P
no 2
∂En (N ) ∂y3o

the sensitivity of the system in (A.2) can be approximated by 6
P
+ν
its sign function [27], [49], [50] as: ∂y3o ∂Wo
j=1 o=1
∂xn ∼ En (N )
 
xn (N ) − xn (N − 1)
= sgn 3 (A.19) ηβ = (B.5)
∂y3o
 ny n 
yo (N ) − y3o (N − 1) 2 
o
∂En (N ) ∂y3o
+ν
P P
6 ∂y3o ∂βol
where sgn(·) is the sign function. l=1 o=1

VOLUME 7, 2019 99345

 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

En (N )
ηµ = " 2 # (B.6) [6] J. Zhang and M. Schroff, ‘‘High-performance micromotor control sys-
nj P
ni P no
 tems,’’ IEEE IECON, Nov. 2003, pp. 347–352.
∂En (N ) ∂xni ∂y3o
+ν
P
6 ∂xni ∂y3o ∂µj
[7] J. Zhang and Q. Jiang, ‘‘Sensorless commutation of micro PMSMs
i=1 j=1 o=1 i for high-performance high-speed applications,’’ in Proc. IEEE ICEMS,
Sep. 2005, pp. 1795–1800.
En (N )
ησ = " 2 # (B.7) [8] Y. H. Chang, T. H. Liu, and C. C. Wu, ‘‘Design and implementation of an
nj P
ni P no
 H∞ controller for a micropermanent-magnet synchronous motor position
∂En (N ) ∂xni ∂y3o
+ν
P
6 ∂xni ∂y3o ∂σ j
control system,’’ IET Electr. Power Appl., vol. 2, no. 1, pp. 8–18, 2008.
i=1 j=1 o=1 i [9] Y.-H. Chang, T.-H. Liu, and C.-C. Wu, ‘‘Novel adjustable micropermanent-
magnet synchronous motor control system without using a rotor-
En (N )
ηα = " 2 # (B.8) position/speed sensor,’’ IEE Proc.-Electr. Power Appl., vol. 153, no. 3,
nj P
ni P no
 pp. 429–438, 2006.
∂En (N ) ∂xni ∂y3o
+ν
P
6 ∂xni ∂y3o ∂α j
[10] Y.-H. Chang, T.-H. Liu, and D.-F. Chen, ‘‘Design and implementation of a
i=1 j=1 o=1 i robust controller for a micro permanent magnet synchronous speed control
systems,’’ in Proc. IEEE ICICIC, Sep. 2007, p. 99.
where ν is a positive constant, then (B.2) can be rewritten as [11] S. E. Lysheski, ‘‘Mini- and microscale closed-loop servodrives with
brushless minimotors and ICs monolithic amplifiers/controllers,’’
En (N + 1) ≈ ν(ηW + ηo + ηβ + ηµ + ησ + ηα ) in Proc. IEEE CDC, Dec. 2002, pp. 3670–3674.
[12] A. Purushotham, S. L. Garverick, C. Edwards, and M. L. Nagy,
νEn (N ) ‘‘A closed-loop micromotor control system,’’ in Proc. IEEE ISCAS,
= " # May 1996, pp. 209–212.
n y n o 2
∂En (N ) ∂y3o

+ν [13] F. F. M. El-Sousy and S. A. Al-Kharj, ‘‘Intelligent hybrid controller for
PP
6 ∂y3 ∂Wj o identification and control of micro permanent-magnet synchronous motor
j=1 o=1
servo drive system using Petri recurrent-fuzzy-neural-network,’’ WSEAS
νEn (N ) Trans. Syst. Control, vol. 9, pp. 336–355, Jul. 2014.
+ " # [14] T.-Y. Chang, T.-H. Liu, and T.-T. Chen, ‘‘Design and implementation of
ny P
no 2
∂En (N ) ∂y3o

+ν an adaptive inverse controller for a micro-permanent magnet synchronous
P
6 ∂y3o ∂Wo motor control system,’’ IET Electr. Power Appl., vol. 3, no. 5, pp. 471–481,
j=1 o=1
2009.
νEn (N ) [15] T.-Y. Chou, T.-H. Liu, and T.-T. Cheng, ‘‘Sensorless micro-permanent
+  ny n  2  magnet synchronous motor control system with a wide adjustable speed
o
∂En (N ) ∂y3o
ν range,’’ IET Electr. Power Appl., vol. 6, no. 2, pp. 62–72, 2012.
P P
6 ∂y3 ∂β l
+
l=1 o=1 o o [16] T. Y. Chou and T. H. Liu, ‘‘Implementation of a motion control system
using micro-permanent magnet synchronous motors,’’ IET Electr. Power
νEn (N ) Appl., vol. 6, no. 6, pp. 362–374, Jul. 2012.
+ " 2 #
nj P
ni P no
 [17] W.-C. Chi and M.-Y. Cheng, ‘‘Implementation of a sliding-mode-based
∂En (N ) ∂xni ∂y3o
+ν
P
6 position sensorless drive for high-speed micro permanent-magnet syn-
∂xni ∂y3o ∂µj chronous motors,’’ ISA Trans., vol. 53, no. 2, pp. 444–453, 2014.
i=1 j=1 o=1 i
[18] F. F. M. El-Sousy and K. A. Abuhasel, ‘‘Nonlinear adaptive backstepping
νEn (N ) control-based dynamic recurrent RBFN uncertainty observer for high-
+ " #
nj P
ni P no
 2 speed micro permanent-magnet synchronous motor drive system,’’ in Proc.
∂En (N ) ∂xni ∂y3o
+ν
P IEEE Energy Convers. Congr. Expo., Portland, OR, USA, Sep. 2018,
6 ∂xni ∂y3o ∂σ j
i=1 j=1 o=1 i pp. 1696–1703.
[19] Y. Li, S. Qiang, X. Zhuang, and O. Kaynak, ‘‘Robust and adaptive back-
νEn (N ) stepping control for nonlinear systems using RBF neural networks,’’ IEEE
+ " 2 # Trans. Neural Netw., vol. 15, no. 3, pp. 693–701, May 2004.
nj P
ni P no

∂En (N ) ∂xni ∂y3o
+ν [20] T. Zhang, S. S. Ge, and C. C. Hang, ‘‘Adaptive neural network control for
P
6 ∂xni ∂y3o ∂α j
i=1 j=1 o=1 i strict-feedback nonlinear systems using backstepping design,’’ Automat-
ica, vol. 36, no. 12, pp. 1835–1846, 2000.
En (N ) En (N ) En (N ) En (N ) [21] J. Y. Choi and J. A. Farrell, ‘‘Adaptive observer backstepping control using
< + + + neural networks,’’ IEEE Trans. Neural Netw., vol. 12, no. 5, pp. 1103–1112,
6 6 6 6
Sep. 2001.
En (N ) En (N ) [22] O. Kuljaca, N. Swamy, F. L. Lewis, and C. M. Kwan, ‘‘Design and imple-
+ + = En (N ) (B.9)
6 6 mentation of industrial neural network controller using backstepping,’’
IEEE Trans. Ind. Electron., vol. 50, no. 1, pp. 193–201, Feb. 2003.
According to (A.1) and (B.9), the convergent ability of the [23] Y. Tan, J. Chang, and H. Tan, ‘‘Adaptive backstepping control and friction
RRBFNN can be guaranteed. compensation for AC servo with inertia and load uncertainties,’’ IEEE
Trans. Ind. Electron., vol. 50, no. 5, pp. 944–952, Oct. 2003.
[24] C.-M. Lin and C.-F. Hsu, ‘‘Recurrent-neural-network-based adaptive-
REFERENCES backstepping control for induction servomotors,’’ IEEE Trans. Ind. Elec-
[1] K. Hameyer and R. Belmans, ‘‘Design of very small electromagnetic and tron., vol. 52, no. 6, pp. 1677–1684, Dec. 2005.
electrostatic micro motors,’’ IEEE Trans. Energy Convers., vol. 14, no. 4, [25] F. J. Lin, P. H. Shieh, and P. H. Chou, ‘‘Robust adaptive backstepping
pp. 1241–1246, Dec. 1999. motion control of linear ultrasonic motors using fuzzy neural network,’’
[2] P. L. Chapman and P. T. Krein, ‘‘Smaller is better? [micromotors and IEEE Trans. Fuzzy Syst., vol. 16, no. 3, pp. 676–692, Jun. 2008.
electric drives],’’ IEEE Ind. Appl. Mag., vol. 9, no. 1, pp. 62–67, Jan. 2003. [26] C. M. Kwan and F. L. Lewis, ‘‘Robust backstepping control of induction
[3] D. P. Arnold and N. Wang, ‘‘Permanent magnets for MEMS,’’ J. Micro- motors using neural networks,’’ IEEE Trans. Neural Netw., vol. 11, no. 5,
electromech. Syst., vol. 18, no. 6, pp. 1255–1266, Dec. 2009. pp. 1178–1187, Sep. 2000.
[4] H. Ishihara, F. Arai, and T. Fukuda, ‘‘Micro mechatronics and micro [27] C.-H. Lee and C.-C. Teng, ‘‘Identification and control of dynamic systems
actuators,’’ IEEE/ASME Trans. Mechatronics, vol. 1, no. 1, pp. 68–79, using recurrent fuzzy neural networks,’’ IEEE Trans. Fuzzy Syst., vol. 8,
Mar. 1996. no. 4, pp. 349–366, Aug. 2000.
[5] T. C. Neugebauer, D. J. Perreault, J. H. Lang, and C. Livermore, ‘‘A six- [28] W.-Y. Wang, T.-G. Leu, and C.-C. Hsu, ‘‘Robust adaptive fuzzy-neural con-
phase multilevel inverter for MEMS electrostatic induction micromotors,’’ trol of nonlinear dynamical systems using generalized projection update
IEEE Trans. Circuits Syst. II, Exp. Briefs, vol. 51, no. 2, pp. 49–56, law and variable structure controller,’’ IEEE Trans. Syst., Man, Cybern. B,
Feb. 2004. Cybern., vol. 31, no. 1, pp. 140–147, Feb. 2001.

99346 VOLUME 7, 2019

F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design
[29] C.-H. Wang, H.-L. Liu, and T.-C. Lin, ‘‘Direct adaptive fuzzy-neural con-
trol with state observer and supervisory controller for unknown nonlinear
dynamical systems,’’ IEEE Trans. Fuzzy Syst., vol. 10, no. 1, pp. 39–49,
Feb. 2002.
[30] O. Castillo and P. Melin, ‘‘Intelligent adaptive model-based control of
robotic dynamic systems with a hybrid fuzzy-neural approach,’’ Appl. Soft
Comput., vol. 3, no. 4, pp. 363–378, Dec. 2003.
[31] C.-J. Lin and C.-H. Chen, ‘‘Nonlinear system control using self-evolving
neural fuzzy inference networks with reinforcement evolutionary learn-
ing,’’ Appl. Soft Comput., vol. 11, no. 8, pp. 5463–5476, 2011.
[32] C.-F. Hsu and K.-H. Cheng, ‘‘Recurrent Fuzzy-Neural approach for non-
linear control using dynamic structure learning scheme,’’ Neurocomputing,
vol. 71, nos. 16–18, pp. 3447–3459, 2008.
[33] H.-C. Lu, M.-H. Chang, and C.-H. Tsai, ‘‘Adaptive self-constructing fuzzy
neural network controller for hardware implementation of an inverted
pendulum system,’’ Appl. Soft Comput., vol. 11, no. 5, pp. 3962–3975,
Jul. 2011.
[34] Y.-S. Lu and J.-S. Chen, ‘‘A self-organizing fuzzy sliding-mode controller
design for a class of nonlinear servo systems,’’ IEEE Trans. Ind. Electron.,
vol. 41, no. 5, pp. 492–502, Oct. 1994.
[35] F. F. M. El-Sousy, ‘‘Adaptive hybrid control system using a recurrent rbfn-
based self-evolving fuzzy-neural-network for PMSM servo drives,’’ Appl.
Soft Comput., vol. 21, no. 8, pp. 509–532, Aug. 2014.
[36] J.-S. R. Jang and C.-T. Sun, ‘‘Functional equivalence between radial basis
function networks and fuzzy inference systems,’’ IEEE Trans. Neural
Netw., vol. 4, no. 1, pp. 156–159, Jan. 1993.
[37] S. Seshagiri and H. K. Khalil, ‘‘Output feedback control of nonlinear
systems using RBF neural networks,’’ IEEE Trans. Neural Netw., vol. 11,
no. 1, pp. 69–79, Jan. 2000.
[38] Y.-K. Yang, T.-Y. Sun, C.-L. Huo, Y.-H. Yu, C.-C. Liu, and C.-H. Tsai,
‘‘A novel self-constructing radial basis function Neural-Fuzzy system,’’
Appl. Soft. Comput., vol. 13, no. 5, pp. 2390–2404, May 2013.
[39] M.-J. Lee and Y.-K. Choi, ‘‘An adaptive neurocontroller using RBFN
for robot manipulators,’’ IEEE Trans. Ind. Electron., vol. 51, no. 3,
pp. 711–717, Jun. 2004.
[40] F. F. M. El-Sousy, ‘‘Adaptive dynamic sliding-mode control system using
recurrent RBFN for high-performance induction motor servo drive,’’ IEEE
Trans. Ind. Informat., vol. 9, no. 4, pp. 1922–1936, Nov. 2013.
[41] J. Fei and T. Wang, ‘‘Adaptive Fuzzy-Neural-network based on RBFNN
control for active power filter,’’ Int. J. Mach. Learn. Cybern., vol. 10, no. 5,
pp. 1139–1150, May 2019.
[42] S. Hou, J. Fei, C. Chen, and Y. Chu, ‘‘Finite-time adaptive fuzzy-
neural-network control of active power filter,’’ IEEE Trans. Power Elec-
tron., vol. 34, no. 10, pp. 10298–10313, Oct. 2019. doi: 10.1109/TPEL.
2019.2893618.
[43] Y. Zhu and J. Fei, ‘‘Disturbance observer based fuzzy sliding mode control
of PV grid connected inverter,’’ IEEE Access, vol. 6, pp. 21202–21211,
2018.
[44] Y. Fang, J. Fei, and D. Cao, ‘‘Adaptive fuzzy-neural fractional-order
current control of active power filter with finite-time sliding controller,’’
Int. J. Fuzzy Syst., vol. 21, no. 5, pp. 1533–1543, 2019. doi: 10.1007/
s40815-019-00648-4.
[45] Z. Zhao, X. He, Z. Ren, and G. Wen, ‘‘Boundary adaptive robust control of
a flexible riser system with input nonlinearities,’’ IEEE Trans. Syst., Man,
Cybern., Syst., to be published. doi: 10.1109/TSMC.2018.2882734.
[46] Z. Zhao, J. Shi, X. Lan, X. Wang, and J. Yang, ‘‘Adaptive neu-
ral network control of a flexible string system with non-symmetric
dead-zone and output constraint,’’ Neurocomputing, vol. 283, pp. 1–8,
Mar. 2018.
[47] Z. Zhao, Y. Liu, W. He, and F. Luo, ‘‘Adaptive boundary control
of an axially moving belt system with high acceleration/deceleration,’’
IET Control Theory Appl., vol. 10, no. 11, pp. 1299–1306, Jul. 2016.
[48] Z. Zhao, X. Wang, C. Zhang, Z. Liu, and J. Yang, ‘‘Neural network
based boundary control of a vibrating string system with input deadzone,’’
Neurocomputing, vol. 275, pp. 1021–1027, Jan. 2018.
[49] S. J. Yoo, Y. H. Choi, and J. B. Park, ‘‘Generalized predictive control based
on self-recurrent wavelet neural network for stable path tracking of mobile
robots: Adaptive learning rates approach,’’ IEEE Trans. Circuits Syst. I,
Reg. Papers, vol. 53, no. 6, pp. 1381–1394, Jun. 2006.
[50] R.-J. Wai and C.-M. Liu, ‘‘Design of dynamic Petri recurrent fuzzy neural
network and its application to path-tracking control of nonholonomic
mobile robot,’’ IEEE Trans. Ind. Electron., vol. 56, no. 7, pp. 2667–2683,
Jul. 2009.
VOLUME 7, 2019
[51] J. J. E. Slotine and W. Li, Applied Nonlinear Control. Upper Saddle River,
NJ, USA: Prentice-Hall, 1991.
[52] K. J. Astrom and B. Wittenmark, Adaptive Control. New York, NY, USA:
Addison-Wesley, 1995.
[53] O. Omidvar and D. L. Elliott, Neural Systems for Control. New York, NY,
USA: Academic, 1997.
[54] L. X. Wang, Adaptive Fuzzy Systems and Control: Design and
Stability Analysis. Englewood Cliffes, NJ, USA: Prentice-Hall,
1994.
[55] B. D. Anderson and J. B. Moore, Optimal Control: Linear Quadratic
Methods. Englewood Cliffs, NJ, USA: Prentice-Hall, 1990.
[56] J.-S. R. Jang, C.-T. Sun, and E. Mizutani, Neuro-Fuzzy and Soft Comput-
ing: A Computational Approach to Learning and Machine Intelligence.
Upper Saddle River, NJ, USA: Prentice-Hall, 1997.
[57] M. Krstic, I. Kanellakopoulos, and P. V. Kokotovic, Nonlinear and Adap-
tive Control Design. New York, NY, USA: Wiley, 1995.
FAYEZ F. M. EL-SOUSY (M’14) received the
B.Sc. degree in electrical engineering from
Menoufia University, Al Minufya, Egypt, in 1988,
and the M.Sc. and Ph.D. degrees in electrical
engineering from Cairo University, Giza, Egypt,
in 1994 and 2000, respectively.
Since 1990, he has been with the Department of
Power Electronics and Energy Conversion, Elec-
tronics Research Institute, Giza, Egypt, where he
is currently a Full Professor. From August 1995 to
June 2003, he was with the Department of Electrical Engineering,
October Six University, Giza Egypt. From April 2004 to February 2007, he
was a Postdoctoral Visiting Researcher with the Graduate School of Infor-
mation Science and Electrical Engineering, Kyushu University, Fukuoka,
Japan. From 2007 to 2010, he was with the Department of Electrical
Engineering, College of Engineering, King Saud University, Riyadh, Saudi
Arabia. From 2010 to 2014, he was with the Department of Electrical
Engineering, College of Engineering, Salman Bin Abdulaziz University,
Al-Kharj, Saudi Arabia. Since 2014, he has been with the Department of
Electrical Engineering, College of Engineering, Prince Sattam bin Abdulaziz
University, Saudi Arabia. His research interests include modeling and control
of motor drives, motion-control systems, wind energy systems, digital signal
processing-based computer control systems, computational intelligent of
power electronics and electric drives, intelligent control theories including
fuzzy logic, neural networks, and wavelets, nonlinear control and optimal
control, and robust control, especially in the intelligent control of Maglev
vehicle transportation systems.
MOHAMED F. EL-NAGGAR received the B.Sc.,
M.Sc., and Ph.D. degrees from Helwan Univer-
sity, Egypt, in 1995, 2002, and 2009, respectively,
all in electrical engineering, where he was with
the Department of Electrical Power and Machines
Engineering, Faculty of Engineering, from 1995
to 2013. Since 2013, he has been an Assistant
Professor with the Department of Electrical Engi-
neering, College of Engineering, Prince Sattam
Bin Abdulaziz University, Saudi Arabia. His main
research interests include power system protection and switchgears, power
transformers operation, testing, and maintenance.
99347
 F. F. M. El-Sousy et al.: Robust Adaptive Neural-Network Backstepping Control Design

MAHMOUD AMIN (S’09–M’12–SM’16) recei- He is the Vice-Chair of the IEEE Computation Intelligence Society, WA
ved the B.Sc. and M.Sc. degrees in electrical Chapter. He is the Editor-in-Chief of the International Journal Electrical
engineering from Helwan University, Egypt, and Electronic Engineering. He is a Regular Reviewer for various IEEE
in 2003 and 2008, respectively, and the Ph.D. Transactions.
degree from Florida International University,
Miami, FL, USA, in 2012. Since 2012, he has
been an Assistant Professor with the Manhattan
College, Riverdale, NY, USA, and Researcher with
the Electronics Research Institute. Since 2018,
he was promoted to an Associate Professor with
Manhattan College. He has published more than 40 articles in professional
journals and conference proceedings. His current research interests include
power electronics in sustainable energy systems, adjustable speed drives, and
smart grid security. He was a recipient of the GSA Scholarly Forum Prize
Paper Award 2010 and the IEEE PES GM 2010 Contest Award. He is Editor
KHALED A. ABUHASEL received the B.Sc. and
of the IEEE TRANSACTIOSN ON ENERGY CONVERSION and the IEEE TRANSACIONS
M.Sc. degrees from the University of Central
ON MAGNETICS.
Florida, Orlando, FL, USA, in 2009 and 2010,
respectively, and the Ph.D. degree from New
AHMED ABU-SIADA (M’07–SM’12) received Mexico State University, Las Cruces, NM, USA,
the B.Sc. and M.Sc. degrees from Ain Shams Uni- in 2012, all in industrial engineering.
versity, Cairo, Egypt, in 1998, and the Ph.D. degree He is currently an Associate Professor with
from Curtin University, Bentley, WA, Australia, the Mechanical Engineering Department, Univer-
in 2004, all in electrical engineering, where he is sity of Bisha, Saudi Arabia. He holds three U.S.
currently an Associate Professor with the Depart- patents, and more than 35 publications in journals
ment of Electrical and Computer Engineering. His and proceeding of very reputable conferences. His research interests include
current research interests include power system optimization, systems engineering, health care systems, intelligent systems,
stability, condition monitoring, power electronics, artificial neural network methodologies, and statistical analysis.
and power quality.

99348 VOLUME 7, 2019